Published in: The Mental Lexicon 3, pp. 149-175, 2008.

The Time-Course of Morphological Constraints:

A Study of Plurals inside Derived Words

Ian Cunnings Harald Clahsen

(University of Essex)

© John Benjamins Publishing Company, 2008. Corresponding Author: Harald Clahsen Department of Linguistics University of Essex Colchester, C04 3SQ, UK email: harald@essex.ac.uk Tel: +44/1206/87-2228 Fax: +44/1206/87-2198

2

Abstract

The avoidance of regular but not irregular plurals inside compounds (e.g. *rats eater vs.

mice eater) has been one of the most widely studied morphological phenomena in the

psycholinguistics literature. To examine whether the constraints that are responsible for this

contrast have any general significance beyond compounding, we investigated derived word

forms containing regular and irregular plurals in two experiments. Experiment 1 was an

offline acceptability judgment task, and experiment 2 measured eye movements during

reading derived words containing regular and irregular plurals and uninflected base nouns.

The results from both experiments show that the constraint against regular plurals inside

compounds generalizes to derived words. We argue that this constraint cannot be reduced to

phonological properties, but is instead morphological in nature. The eye-movement data

provide detailed information on the time-course of processing derived word forms indicating

that early stages of processing are affected by a general constraint that disallows inflected

words from feeding derivational processes, and that the more specific constraint against

regular plurals comes in at a subsequent later stage of processing. We argue that these results

are consistent with stage-based models of language processing.

3

1. Introduction

Morphological systems are constrained in how inflectional, derivational, and compounding

processes may interact with each other. There is, for example, a general tendency for

inflectional suffixes to attach outside derivational ones, which is captured by Greenberg’s

(1966: 93) Universal 28: ‘If both the derivation and inflection follow the root, or they both

precede the root, the derivation is always between the root and the inflection’. Thus in

English, forms such as ducklings with the plural suffix appearing after the derivational suffix

are grammatically well-formed but not forms with the reverse order (*ducksling).

Linguists have analyzed the interaction between different morphological processes in terms

of affix-specific selectional restrictions (see e.g. Fabb 1988, Plag 1996, 1999, 2002) and

other kinds of more general constraints (of a semantic, phonological, and morphological

nature) on how inflectional, derivational, and compounding processes may feed each other

(see e.g. Kiparsky 1982, Wiese 1996, Anderson 1992, Stump 2001). In English, the

observation that the denominal suffix –y never attaches to an already suffixed word (Fabb

1988), as in *duckling-y, is an example of an affix-specific selectional restriction. A well-

known example of a more general morphological constraint states that regularly inflected

forms such as –s plurals are blocked from productive derivational processes and from

compounding (Kiparsky 1982), hence the ungrammaticality of forms such as *ducks-less or

*ducks eater.

From a psycholinguistic perspective, the question arises as to how these kinds of constraints

affect the processing of morphologically complex word forms in real time. Specifically, we

ask at which point in time different kinds of constraints become operative. Previous

psycholinguistic research investigating the interaction of different morphological processes

has focused on the feeding relationships between inflection and compounding, the so-called

plurals-in-compounds effect (Senghas, Kim & Pinker. 2005). Several off-line rating and

4

production studies (Gordon 1985, Haskell, MacDonald & Seidenberg 2003, Berent & Pinker

2007, among others) have found a response pattern indicating a general dispreference for

compounds containing plural non-heads (irrespective of regularity) in comparison to those

containing singulars, and an additional dispreference for compounds containing regular

plurals. In judgment tasks, for example, both regular and irregular plurals have been found to

be less acceptable inside compounds than singulars (fleas/lice remover vs. flea/louse

remover) and regular plurals have been found to be less acceptable than irregular plurals

(fleas remover vs. lice remover). In an eye-movement study, Cunnings and Clahsen (2007)

found that early eye-movement measures showed enhanced reading times for compounds

containing regular plural non-heads only, whilst later measures revealed an additional effect

for irregular plural versus singular non-head nouns. Whilst the general preference for

singular non-heads has been ascribed to a semantic constraint against compound-internal

modifiers with plural meaning (Haskell et al. 2003, Berent & Pinker 2007), the interpretation

of the contrast between regular and irregular plural modifiers is controversial. Some authors

have argued that this contrast is due to the phonological properties of regular plural nouns,

such that there is a specific dislike of compound-internal modifiers that end in s or z (Haskell

et al. 2003), or a result of the higher phonological overlap between a regular plural and its

corresponding singular form than between an irregular plural and its corresponding singular

form (Buck-Gengler, Menn & Healy 2004). Others have argued that this contrast is due to a

morphological constraint against regular plural modifiers inside compounds (see e.g.

Aronoff 1976, Kiparsky 1982, Di Sciullo & Williams 1987, Borer 1988, Pinker 1999).

In contrast to compounding and inflection, there is very little experimental research on the

interaction of inflectional and derivational processes. Marshall and van der Lely (2007)

performed an elicited production experiment in which children with Specific Language

Impairment (SLI) were found to sometimes include the inflectional suffix –s inside a

5

denominal adjective with –y when presented with a plural stimulus (i.e. with a picture

showing multiple objects of a given kind), producing forms such as *stars-y, whereas

unimpaired children almost never did so, even when prompted with a plural stimulus. This

finding was taken to indicate that children with SLI treat plural forms such as stars as

unanalyzed wholes which as such may feed derivation. Apart from this one study, which

crucially did not compare regular to irregular plurals inside derived words, we are not aware

of any other experimental investigation on how inflectional and derivational processes

interact.

The present study aims to fill this empirical gap. We report results from two experiments, an

acceptability judgment task examining whether inflected forms inside derivations exhibit the

same contrasts between regulars and irregulars and between singulars and plurals that have

been found for compound-internal inflected forms, and an eye-movement experiment

examining the processes involved in the on-line comprehension of inflections inside

derivations.

2. Constraints on inflections inside derivations in English

Derivational affixes can be combined with each other in English, subject to a number of

constraints. Whilst, for example, combinations of the suffixes –less and –ness, and –ic and

-ity are grammatically well-formed (e.g. atom-less-ness, atom-ic-ity), the combination

-less+–ity is not possible (compare *atom-less-ity with atom-less-ness). These combinatorial

properties have been described in terms of affix-specific selectional restrictions (Fabb 1988,

Plag 1996, 1999, 2002) akin to the selectional restrictions of verbs, adjectives and other

kinds of lexical items. For example, the affix –ness may select derivational stems with –less

(yielding atom-less-ness), the affix –ity stems with –ic (yielding atom-ic-ity), but not the

6

other way round. In this way, selectional restrictions may determine the possible and

impossible combinations of different affixes.

Apart from these kinds of affix-specific restrictions, derivational processes are also subject to

a general restriction that permits uninflected and disprefers inflected forms inside derived

forms yielding contrasts such as flealess vs. *fleasless, rocky vs. *rocksy, or louseless vs.

?liceless. Given realization-based models of morphology, such as Anderson (1992),

Matthews (1991) and Stump (2001), these contrasts can be derived from a Category

Constraint on the kinds of morphological types or categories that may enter derivational

processes. Realization-based models distinguish between rules that define derivational stem

entries from those that define inflected word forms. Derivational rules map one stem entry

onto another entry. A derived entry (e.g. duckling) may provide the input to subsequent

derivational rules (e.g. duckling-less), or may provide the base for the application of

inflectional rules (duckling-s). In contrast, inflectional rules are simple feature-form

mappings that specify the form that ‘realizes’ or ‘spells out’ a particular set of features. Sets

of inflectional rules are partitioned into ordered blocks or paradigms with the forms defined

by each rule block being passed to the next block without at any point being represented in

lexical entries. This implies that the outputs of inflectional rules (= word forms) do not

provide the input required for derivational processes (= stem entries), hence the

ungrammaticality of forms such as *ducksling. Thus, from the perspective of realization-

based models of morphology, the Category Constraint is a consequence of the different

nature of inflectional and derivational processes.

In addition to the preference of uninflected over inflected forms inside derivations, native

speakers of English when informally asked to make a choice between a regular and an

irregular plural inside a derived form, e.g. *fleasless vs. ?liceless, tend to find the latter more

acceptable than the former. This could mean that derivational processes are subject to the

7

same constraint that blocks regularly inflected forms from appearing inside compounds;

compare, for example, *fleas eater vs. ?lice eater. As mentioned above, the nature of this

constraint is controversial. Some researchers (Buck-Gengler et al. 2004, Haskell et al. 2003)

have argued that the contrast between regular and irregular plurals inside compounds is due

to the surface form properties of regular plurals, e.g. the dispreference of compound-internal

modifiers that end in s or z (Haskell et al. 2003). Note, however, that English has a number

of uninflected bare nouns that share the phonological characteristics of regular plurals, e.g.

fox, box, hose, tax, etc., and these have been found to be fully acceptable inside compounds

(Berent & Pinker 2007). The same contrast seems to hold for derived words. Compare, for

example, *dogsless vs. wolfless and foxless, of which the latter two appear to be equally

acceptable even though foxless has the same stem-final phonology as the ungrammatical

regular plural (dogs) inside the derived form. Given these data, a purely phonological

account of the contrast between regular and irregular plurals inside compounds and derived

forms appears to be difficult to maintain (see Cunnings & Clahsen 2007 for further

discussion).

More commonly, the interaction of regular and irregular inflection with compounding and

derivation has been analyzed in morphological terms. Kiparsky (1982) proposed a theory of

morphological representation that includes a Level Ordering component which orders

irregular inflection before, and regular inflection after, compounding and (productive)

derivation. This component prevents regular inflectional affixes from appearing inside

compounds or inside productive derived forms. Several subsequent accounts have attempted

to reformulate the descriptive insights of Kiparsky’s model in terms of other distinctions (see

e.g. Di Sciullo and Williams 1987, Borer 1988, Wiese 1996, Pinker 1999). For derivational

morphology, Hay (2002, 2003) argued that the possible and impossible combinations of

derivational suffixes in English are determined by a parsability constraint according to which

8

‘an affix that can be easily parsed out should not occur inside an affix that cannot’ (Hay

2002: 528). Consequently, highly decomposable suffixes tend to occur towards the periphery

of words rather than inside less decomposable ones. Whilst the notion of parsability was

originally proposed to explain the combinatorial properties of derivational suffixes, it may in

principle be extended to account for the contrast between regular and irregular plurals inside

derived forms in English. Regular plural forms are highly decomposable and given the

parsability constraint less likely to occur in a derived word than irregular plurals which are

typically suppletive forms and not decomposable. Hence ratsless is intuitively less

acceptable than miceless.

Taken together, we hypothesize that the interaction of inflectional and derivational processes

is subject to two different constraints, a general Category Constraint that disprefers all kinds

of inflected forms inside derived forms and an additional (morphological) Structure

Constraint that prevents regularly inflected forms (which are morphologically structured and

decomposable) from feeding any further word-formation processes. Given these constraints,

we would expect to find differences in acceptability ratings of derived forms such as (i)

flealess/louseless, (ii) liceless and (iii) fleasless. Forms with uninflected bare stems (i) should

receive the highest acceptability ratings, because they do not violate any of the constraints.

Derived forms such as (ii) violate the Category Constraint but not the Structure Constraint

and may therefore be marginally acceptable. Finally, forms such as fleasless violate both

constraints and should therefore be the least acceptable. Whilst introspectively these

distinctions appear to be correct, the purpose of Experiment 1 was to examine these

intuitions more systematically in an acceptability judgment task of derived words containing

base nouns that were the singular and plural forms of nouns that take regular and irregular

inflections.

9

3. Morphological constraints during language processing

Psycholinguists attempt to specify at which point in time particular sources of information

become available during on-line language comprehension. Broadly speaking, we can

distinguish between serial and parallel-interactive models of language processing. Serial

models claim that language comprehension involves a sequence of stages and that the

computation of structural (morpho-syntactic) information precedes the computation of

semantic and pragmatic information (e.g. Frazier & Fodor 1978, Rayner, Carlson & Frazier

1983, McElree & Griffith 1998, Frazier & Clifton 1996, Binder, Duffy & Rayner 2001). The

alternative parallel-interactive view holds that language comprehension at any given point in

time is affected by all relevant sources of information (e.g. Altmann, Garnham & Dennis

1992, Thornton, Gil & MacDonald 1998, Spivey, Fitneva, Tabor & Ajmani 2002).

Given that the interaction of inflectional and derivational processes in morphologically

complex words is subject to constraints and that these constraints rely on different kinds of

information, time-course issues also arise for the processing of derived word forms. The

Category Constraint, for example, excludes certain kinds of objects (inflected words) from

feeding derivational processes, whilst the Structure Constraint specifies their internal

structure, ruling out regularly inflected combinatorial forms from appearing inside derived

word forms. Given these constraints, the question arises as to at which point in time they

affect the processing of a derived word form. Does the Category Constraint become available

earlier in time than the Structure Constraint, or vice versa, or are both constraints employed

in parallel? To our knowledge, the interaction of inflectional and derivational processes

during on-line processing has previously not been investigated. However, by extrapolating

from previous sentence-processing research, we can think of two possibilities of how these

constraints may affect the on-line processing of derived words. One possibility is that all

relevant constraints apply in parallel and that corresponding effects are seen at both early and

10

late stages of processing. This would fit in with parallel-interactive models of sentence

processing in which all relevant information sources are used the moment they are received

as input (Spivey, Fitneva, Tabor & Ajmani 2002; 208). An alternative possibility is that the

two constraints on derived word forms become available at different points in time during

processing, which would be consistent with stage-based models of sentence processing.

Friederici (2002), for example, proposed a model according to which word-category

information is processed earlier than other aspects of morpho-syntactic structure. If this

extends to morphologically complex words, we may find effects of the Category Constraint

at earlier stages of processing than effects of the Structure Constraint. These different

possibilities were examined in Experiment 2 by recording eye-movements during reading, a

technique that provides a rich source of data on moment-to-moment language processing

(Rayner 1998).

4. Experiment 1: Acceptability judgments

The purpose of this experiment was to examine whether forms such as (i) flealess/louseless,

(ii) liceless, and (iii) fleasless show the same distinctions in acceptability ratings that have

previously been found for inflected and non-inflected forms inside compounds. Moreover, as

pointed out above, the nature of the plurals-in-compounds effect is controversial, with some

authors claiming that it is due to a phonological constraint (e.g. against modifiers ending in

s/z) and others arguing that it is morphological in nature (e.g. banning compound-internal

inflected words with a combinatorial structure). We therefore adopted the acceptability rating

task from previous studies of compounding (e.g. Haskell et al. 2003, Senghas et al. 2005,

Berent & Pinker 2007) to examine derived words such as those in (i) to (iii), and to test

whether any dissociation between regular and irregular plurals inside derived words is best

characterized in morphological or phonological terms.

11

Method

Participants

40 native English speakers (11 males, mean age 24.8) from south-east England voluntarily

took park in this experiment.

Materials

All experimental items were derived words constructed by attaching a single derivational

suffix, adjective-forming -less and adverb-forming -wise, to a base noun. There were two

sets of experimental items treated separately for analysis purposes. Experimental item set 1

was constructed to examine the acceptability of derived words containing the singular and

plural forms of base nouns that take regular and irregular inflections. Nine nouns that take

irregular plural forms and nine nouns that take the regular plural –s were chosen, which were

as closely related in meaning as possible (e.g. irregular goose/geese was matched with

regular duck/ducks). The singular forms of these nouns were matched for frequency, length

in letters and number of syllables, using the CELEX Lexical Database (Baayen, Piepenbrock

& van Rijn 1993). For the frequency matching, the word-form frequency of the singular

forms of the nouns that take irregular plurals was matched with the lemma frequency of the

nouns that take regular plurals. This was done to do justice to the linguistic differences

between regular and irregular plurals in English. Irregular plurals are largely suppletive

forms whereas regular -s plurals are fully transparent forms with a segmentable ending and

no stem changes. One would therefore expect that the recognition of irregular plural forms is

mainly affected by word-form frequency, whereas the recognition of regular plurals should

be affected by lemma/stem frequency; see Sonnenstuhl & Huth (2002) for evidence from

German noun plurals supporting this contrast. Overall, the items were tightly controlled for

these three matched variables. The mean frequency per million of the irregular and regular

12

nouns was 207.67 and 204.67 respectively, mean length in letters for the irregular and

regular nouns was 4.33 and 4.22 respectively, and mean number of syllables for the irregular

and regular nouns was 1.11 and 1.33 respectively. A series of between-items t tests indicated

that there were no significant differences between the singular forms of the irregular and

regular nouns, for any of the three matched variables (all t < 1.33, all p > .208). A full list of

items can be found in Appendix A. Both the singular and plural forms of each noun were

tested with the two derivational suffixes -less and -wise, creating a total of 72 items in four

conditions (with 18 items per condition): (i) singular forms of nouns that take irregular

plurals (singular/IRREGULAR, e.g. gouseless), (ii) irregular plural forms

(plural/IRREGULAR, e.g. geeseless), (iii) singular forms of nouns that take regular plurals

(singular/REGULAR, e.g. duckless), (iv) regular plural forms (plural/REGULAR, e.g.

ducksless).

Experimental item set 2 was constructed to examine the effects of base noun phonology on

the acceptability of derived words. All items in experimental item set 2 were singular nouns.

Eight nouns that phonologically resemble regular plurals and eight nouns that are

phonologically distinct from regular plurals were chosen, which were as closely related in

meaning as possible (e.g. box was matched with crate). These nouns were additionally

matched for lemma frequency per million, length in letters and number of syllables. The

mean frequency, mean length in letters and mean number of syllables of the nouns that

phonologically resembled regular plurals was 32.13, 4.25 and 1.25 respectively, whilst the

mean frequency, mean length in letters and mean number of syllables of the nouns that were

phonologically distinct from regular plurals was 33.88, 4.13 and 1.00 respectively. A series

of between-items t tests revealed no significant differences between the two types of noun,

for any of the three matched variables (all t < 1.53, all p > .148). A full list of items can be

found in Appendix B. As with experimental item set 1, the two derivational suffixes –less

13

and –wise were attached to the base nouns of experimental item set 2. As only the derived

forms containing the singular forms of each noun were tested, experimental item set 2

consisted of a total of 32 derived words in two conditions (with 16 items per condition): (i)

singular forms of nouns that end in codas phonologically identical to regular plural forms

(phon./SIM, e.g. boxless), (ii) singular forms of nouns that end in codas phonologically

dissimilar to regular plural forms (phon./DIS, e.g. crateless).

In addition to the two sets of experimental items, 40 filler items were constructed, all of

which were morphologically complex words containing a variety of affixes and affix

combinations. 10 of the filler items were acceptable complex words using attested affix

combinations in the language (e.g. fatherly, undriveable), whilst a further 10 fillers were

unacceptable complex words, using unattested affix combinations (e.g. oddnessity,

strongity). An additional 10 fillers were constructed which were unacceptable complex

words containing various suffixes, all of which were attached to nouns in their plural form

(e.g. parentshood, apesdom). A final 10 unacceptable filler items were constructed by

attaching various deverbal suffixes to verbs which appeared in their past tense form (e.g.

sailedable, drover).

A series of short two-sentence context paragraphs were constructed within which all

experimental and filler derived words were inserted (see (1)). The derived words always

appeared in the second sentence of the context paragraphs, in bold capital letters to ensure

their salience within the texts. For experimental item set 1, context paragraphs were

constructed such that a derived word from each of the four conditions could be placed

felicitously into an identical context paragraph. The same procedure was adopted for the two

conditions of experimental item set 2.

14

(1) a. Experimental item set 1, -less form:

Following recent food scares regarding British meat, the Government has advised

farmers to slaughter much of their livestock. One local farmer has been left

completely OX(EN)/PIG(S)-LESS since the cull was recommended.

b. Experimental item set 1, -wise form:

Many dog owners have problems keeping their pets free from mites and other itchy

pests. LOUSE(LICE)/FLEA(S)-WISE, our dog has luckily never had such

problems.

c. Experimental item set 2, -less form:

My sister recently moved, and was annoyed with all the storage containers cluttering

up her new home. Although the house has been completely BOX/CRATE-LESS for

a week now, she’s still having trouble settling in.

d. Experimental item set 2, -wise form:

Wild animals always seem to terrorise local farmers by eating most of their livestock.

WOLF/FOX-WISE, there has been a particularly worrying amount of trouble this

year.

Procedure

Four lists were created such that each participant saw each context paragraph only once with

each list containing 34 experimental items and 40 fillers. The experimental and filler items

were presented in the same pseudo-randomized order for each list, with no experimental

items from the same condition appearing adjacent to each other. Participants were instructed

to simply read each context paragraph and then rate the acceptability of the derived word

only, on a scale from 1 (highly unacceptable) to 7 (highly acceptable). Each list was

completed by the same number of participants.

15

Results

The mean acceptability ratings (and standard deviations) for the derived words of the two

experimental item sets are shown in Table 1.

//INSERT TABLE 1 ABOUT HERE//

The results for experimental item set 1 indicate that derived words containing singular base

nouns received high acceptability ratings. By contrast, derived words containing regular

plurals were rated highly unacceptable and derived words containing irregular plurals

marginally acceptable. A two-way ANOVA with Regularity (irregular/regular) and Plurality

(singular/plural) as independent variables on the mean acceptability scores for experimental

item set 1 revealed significant main effects of Regularity (F1(1, 39) = 28.43, p < .001; F2(1,

68) = 13.60, p < .001) and Plurality (F1(1, 39) = 101.06, p < .001; F2(1, 68) = 76.66, p <

.001), and most importantly, a significant Regularity-by-Plurality interaction (F1(1, 39) =

21.03, p < .001; F2(1, 68) = 7.78, p = .007). Subsequent pairwise comparisons revealed that

there were no significant differences between the acceptability of derived words containing

the two different types of singular nouns (sg./IRR vs. sg./REG: t1(39) < 1, p = .530; t2(34) <

1, p = .587). Derived words containing irregular plurals were, however, rated significantly

less acceptable than derived words containing either type of singular noun (pl./IRR vs.

sg./REG: t1(39) = 6.03, p < .001; t2(34) = 3.65, p = .001; pl./IRR vs. sg./IRR: t1(39) = 5.99, p

< .001; t2(34) = 3.71, p = .001). Derived words containing regular plurals were also rated

significantly less acceptable than derived words containing either type of singular base

(pl./REG vs. sg./REG: t1(38) = 9.37, p < .001; t2(34) = 9.71, p < .001; pl./REG vs. sg./IRR:

t1(39) = 9.48, p < .001; t2(34) = 8.65, p < .001). Additionally, derived words containing

regular plurals were rated significantly less acceptable than derived words containing

16

irregular plurals (pl./REG vs. pl./IRR: t1(39) = 5.61, p < .001; t2(34) = 5.65, p < .001). These

results reveal a gradual pattern of responses with both regular and irregular plural base nouns

being less acceptable inside derived words than uninflected bare stems. Derived words

containing regular plurals are also less acceptable than derived words containing irregular

plurals.

As shown in Table 1, experimental item set 2 yielded almost the same acceptability ratings

of 5.08 for the phonologically similar condition and 5.07 for the phonologically dissimilar

condition (phon./SIM vs. phon./DIS: (t1(39) < 1, p = .974; t2(30) < 1, p = .994), indicating

that a singular noun that ends in the same segment as a regular plural did not have any

adverse effect on participants’ acceptability judgments.

Discussion

These results are parallel to what has previously been found for compounds. Recall that the

same contrasts we found for derived forms (with both regular and irregular plurals being less

acceptable inside derived words than singulars, and regular plurals being also less acceptable

than irregular plurals) have previously been obtained for inflected and non-inflected forms

inside compounds in a number of studies (Haskell et al. 2003, Senghas et al. 2005, Cunnings

& Clahsen 2007). Likewise, our finding that the phonological properties of singular base

nouns (in particular /s/ final codas) had no reliable effect on the acceptability of the derived

words is also parallel to Berent and Pinker’s (2007) observation that compounds such as fox

chaser are rated as being fully acceptable.

These findings could mean that compounding and productive derivational processes such as

–less and –wise suffixation are subject to the same constraints. For plurals inside compounds,

three constraints have been proposed. One proposal was a phonological constraint according

to which s/z-final words are dispreferred inside compounds (Haskell et al. 2003). Berent and

17

Pinker (2007) provided evidence from compounds against such a constraint, and our results

clearly show that the low acceptability ratings of regular in comparison to irregular plurals

inside derived forms with -less and -wise cannot be explained in phonological terms. More

commonly, the dislike of regular plurals inside compounds has been attributed to a constraint

of morphological structure (see e.g. Kiparsky 1982, Di Sciullo and Williams 1987, Wiese

1996, among others). Our results indicate that this constraint also affects derivation, thus

confirming Kiparsky’s (1982) original insight that regular inflection is ordered after all other

word-formation processes and can therefore not feed either compounding or derivation.

A further constraint has been proposed for compounds to capture the fact that compounds

containing singular nouns as non-heads are preferred over plural forms of any kind,

irrespective of regularity. Haskell et al. (2003) argued that this contrast is due to a semantic

constraint against compound-internal modifiers with plural meaning. Specifically, the

semantics of compounds requires that the relationship between the referents of the two

elements of a compound is interpreted such that the non-head element of a compound

typically refers to a kind, not an individual. For example, a garden hose can be used for

gardens in general, not for a particular kind or number of gardens. Given that in English, a

singular noun form is identical to a bare nominal stem, the form which is not explicitly

marked for NUMBER is more acceptable inside a compound than a plural form. Moreover,

plural NUMBER semantics has been shown to affect the extent to which a plural noun is

admissible inside a compound. In compounds such as publications catalogue, for example,

the plural non-head refers to multiple distinct entities (i.e. publications) within a

heterogeneous collection of individuals, and in such cases even a regular plural form can be

fully acceptable (see Alegre & Gordon 1996, 1999; Pinker 1999). These considerations do

not seem to apply to derived words. Compounds consist of a modifier and a head and may

therefore be affected by a semantic constraint that specifies the relationship between

18

different referents of the modifier and the head of a compound. Derived words, however,

consist of a base and one or more affixes. Derivational affixes do not have referents, and the

base of a derived word does not constitute a ‘modifier’ of any kind. Consequently, the

semantic constraint as proposed for compounds does not apply to derived words. Instead,

derivational affixes are subject to selectional restrictions, some of which are affix-specific

and others more general. As mentioned above, one such general constraint restricts the kinds

of morphological types or categories that may enter derivational processes to stems.

According to this Category Constraint, uninflected stems are preferred and all kinds of

inflected words dispreferred inside derived forms yielding contrasts such as flealess vs.

*fleasless or louseless vs. ?liceless.

Summarizing, we propose that derived words are subject to a Category Constraint and to the

same (morphological) Structure Constraint that applies to compounds. Derived words

containing uninflected stems (flealess) do not violate any of these constraints and are fully

acceptable, whilst those containing regular plurals violate both constraints and are

ungrammatical (*fleasless), and derived words containing irregular plurals violate the

Category but not the Structure Constraint and as such are marginally acceptable (?liceless).

In this way, the two constraints account for the performance pattern seen in the present

experiment.

5. Experiment 2: Eye-movements during reading derived words

The aim of this experiment was to examine the time-course of the constraints on derived

word forms investigated in the acceptability judgment task reported above. Participants read

a series of paragraphs containing derived words similar to those tested in Experiment 1,

whilst their eye-movements were monitored. From the perspective of stage-based models of

language processing, one would expect that violations of the Category Constraint (against

19

inflected words inside derivations) should lead to increased reading times from the earliest

possible measures, whereas violations of the Structure Constraint (against regular inflection)

should be recognized later (see section 3 above). Alternatively, we may find that the two

constraints affect reading-time measures in parallel, a result which would be more

compatible with parallel-interactive models of language processing.

Method

Participants

Twenty-four native English speakers (12 males, mean age 21.8) with normal or corrected to

normal vision from the University of Essex student community were paid to participant in

Experiment 2, none of whom took part in Experiment 1.

Materials

Experimental items consisted of 54 novel derived words in three conditions, containing (i)

irregular plurals (e.g. oxenless), (ii) regular plurals (animalsless), (iii) singular nouns

(horseless). Unlike in experiment 1, we used different base nouns for these three conditions

as it is impossible to match the singular and plural forms of the same base noun (e.g. rat and

rats) for length, a variable that has been shown to have an immediate effect on processing

times during reading (Rayner 1998). Instead, nine irregular plural nouns were matched with

nine regular plural nouns and nine singular nouns. Although the base nouns differed between

conditions, they were matched as closely as possible for a number of lexical variables that

may potentially influence reading times. These included frequency, length in letters and

number of syllables using the CELEX lexical database. As in experiment 1, word-form

frequencies were used for the irregular plurals items, whilst lemma frequencies were used for

the regular plural and singular items, in accordance with the linguistic differences between

20

regular and irregular plural forms in English. The mean frequencies per million were 238.22,

239.78 and 235.78 for the irregular plural, regular plural and singular conditions

respectively. The mean length in letters was 4.67, 4.67 and 4.33, and the mean number of

syllables was 1.33, 1.22 and 1.33 for the irregular plural, regular plural and singular

conditions respectively. A series of between-items one-way ANOVAs revealed that there

were no significant differences between the three conditions for any of the three matched

variables (all Fs < 1, all ps > .791). Appendix B contains frequency and length information

for all nouns used. In addition, the items were matched with respect to concreteness and

familiarity, two variables which correlate with reading times (Juhasz & Rayner 2003). Two

one-way ANOVAs revealed that the base nouns in each condition did not differ in either

concreteness or familiarity (both Fs <1, both ps > .790), in terms of the measures provided

by the MRC Psycholinguistic Database1 (Wilson 1988). Furthermore, there were also no

significant differences between conditions in terms of the base nouns’ mean lexical decision

times and naming latencies2 (both Fs < 1, both ps > .397), as provided by the English

Lexicon Project (Balota, Cortese, Hutchison, Neely, Nelson, Simpson & Treiman 2002). We

thus maintain that the base nouns between conditions were very closely matched for a

number of variables that could potentially influence reading times.

Each noun was affixed with two derivational suffixes, adjective-forming –less and adverb-

forming suffix -wise, creating a total of 54 experimental items, with 18 items per condition

(e.g. oxenless, animalsless, horseless and oxenwise, animalswise, horsewise). Of all the

critical derived words we tested, only one occurs in CELEX (faceless), with a low word-

form frequency of 2 per million. Furthermore, none of the -less or -wise forms containing

irregular or regular plural base nouns were attested in the much larger (100 million word)

British National Corpus (BNC), even though compounds containing irregular plural non-

heads, such as lice infestation, and even regular plurals, such as buildings inspector, are

21

found in the corpus. Of the -less and -wise forms with singular base nouns, only one token

with -wise occurred in the entire corpus (facewise) and six forms with -less (fishless, dogless,

handless, husbandless, faceless), all of which had very low token frequencies of close to zero

per million in the BNC. Thus, although we cannot rule out the possibility that participants

may have previously encountered some of the derived words, it is maintained that all are

sufficiently low in frequency to minimize potential whole-word frequency effects.

To ensure a felicitous reading of the derived words, short three-sentence context paragraphs

were created for all experimental and filler items. For each triplet of items (e.g. oxenless,

animalsless and horseless formed one triplet), three different context paragraphs were

constructed, within which each critical item could be placed felicitously. Each context

paragraph began with a sentence that introduced a character and continued with a second

sentence that always included the critical derived word. Each derived word appeared towards

the middle of the second sentence, and was always preceded by a pre-target word of between

6 and 12 letters long. The pre-target word was chosen to be of this length to aid the

probability of the target word receiving a progressive first fixation during its first pass. The

third sentence was a simple wrap-up sentence to soak-up any end of trial effects. Each

sentence of each context paragraph took up one line onscreen. An example of an

experimental triplet of context paragraphs is shown in (2):

(2) a. Example context 1:

Our dog was pest infested until we treated him with a special shampoo.

He's since been completely LICE/MITES/FLEALESS and has stopped itching all over.

If he has similar problems in the future we now know what to do.

b. Example context 2:

Our dog Max often has pest problems and is always scratching himself.

22

However, he's been totally LICE/MITES/FLEALESS since we bought a pest shampoo.

He hasn't scratched himself since we used the shampoo last Wednesday.

c. Example context 3:

Our cat is usually covered in pests as she loves playing in our bins.

However, she's been practically LICE/MITES/FLEALESS since we removed the bins.

Now that she can't get to the trash, we hope she won't itch as much.

In addition to the experimental items, 60 filler texts comprised of a variety of different

syntactic constructions were created. Each filler text took up to three lines onscreen, half of

which contained two sentences, and half of which contained three sentences.

To ensure that the critical items displayed the intended range of acceptability, the 54

experimental context paragraphs were pre-tested with 21 native English speakers (6 males,

mean age 28.1) from the University of Essex student community, none of whom took part in

any of the other experiments. In this pre-test, the 54 critical items were interspersed with 40

fillers that contained a variety of acceptable and unacceptable derived words. Three pseudo-

randomized lists were created such that no critical items from the same condition appeared

adjacent to each other, and each list was distributed to the same number of participants, who

rated the acceptability of the derived words only on a scale from 1 (unacceptable) to 7 (fully

acceptable). The mean acceptability ratings of the experimental items were as follows.

Derived words containing singular base nouns received a high overall mean acceptability

rating of 4.67, whilst those containing regular plural base nouns received a low mean

acceptability rating of 1.92. Derived words containing irregular plural base nouns received a

marginally acceptable mean rating of 3.47. A one-way ANOVA revealed a significant main

effect of base noun type (F1(2, 40) = 74.52, p < .001; F2(2, 51) = 148.10, p < .001), and

subsequent pairwise comparisons revealed significant differences between each of the three

23

conditions (4.67 vs. 3.47: t1(20) = 6.49, p < .001; t2(34) = 6.83 < .001; 4.67 vs. 1.92: t1(20) =

9.95, p < .001; t2(34) = 16.43, p < .001; 3.47 vs. 1.92: t1(20) = 7.52, p < .001; t2(34) = 11.49,

p < .001). These pre-test results confirm the findings of Experiment 1 for the materials to be

used in the eye-movement experiment.

Procedure

For the main experiment, three lists were constructed such that each participant saw each of

the 54 experimental items, and each context paragraph only once. The 54 experimental items

were pseudo-randomised with the 60 fillers such that no two experimental items from the

same condition appeared adjacent to each other. In addition, the ordering of the three

versions of the experiment were reversed, making a total of six experimental lists, which

were all completed by the same number of participants, with all derived word and context

paragraph combinations being seen by the same number of participants across lists.

Participants were instructed to read the sentences onscreen at their normal reading rate, and

press a button once completed. To ensure that participants properly attended to the sentences,

simple content questions requiring a yes/no push-button response were displayed after

approximately 20% of trials. Participants had few problems with the content questions,

giving correct responses with a mean accuracy of 89.2% (SD 7.82), indicating that they paid

attention to the content of the sentences.

Eye-movements were recorded using the head-mounted EYELINK II system from SR

Research Ltd., Canada. Stimuli were presented onscreen 80cm from the participants’ eyes,

and at this viewing distance one onscreen character subtended approximately one degree of

visual angle. The eye-tracker records participants’ eye-movements via two cameras mounted

on a headband placed over the participants’ head at a sampling rate of 500Hz, with spatial

accuracy better than 0.5 degrees. Participants’ head movements are automatically

24

compensated for via a third camera mounted in the centre of the headband, which tracks the

position of four LEDs mounted on the corners of the computer screen. Any head movement

is detected as movement of the LEDs, and is automatically compensated for online. Whilst

participant vision was binocular, the eye-movement record was recorded from the right eye

only. Before presentation of the main experiment, 5 practice trials were displayed onscreen

to familiarize participants with the experimental procedure. Participants were instructed to sit

still whilst reading, and were allowed breaks in-between trials if they so wished. The entire

experiment lasted approximately 40 minutes.

Data Analysis

Reading time measures were calculated for two regions of text. The pre-target region was

defined as the word directly preceding the derived word, plus half a letter space either side,

and the derived word region was defined as the derived word itself, plus half a letter space

either side. Whilst the derived word is the primary region of interest, reading times were also

calculated for the word directly preceding it to capture any potential effects of parafoveal

processing.

Five reading-time measures were calculated to provide a rich source of information

regarding the time-course of processing. First fixation duration refers to the duration of the

first fixation within a target region. Gaze duration is calculated by summing the duration of

all fixations, starting with the first fixation when the eyes first enter a region from the left, up

until the eyes first leave that region, either to the left or the right. Both of these measures are

believed to index the early stages of processing that accompany and immediately follow

lexical access, although gaze duration times may reflect later processing stages than first-

fixation durations, because large regions of text (including large words) often obtain more

than one fixation during their first pass. The third measure calculated was regression path

25

duration. This is calculated by summing the duration of all fixations, starting with the first

fixation when a region is entered from the left, up until but not including the first fixation

when the target region is first exited to the right. As this measure includes the duration of any

regressive fixations, regression path durations are usually longer than first fixation and gaze

durations, and can be seen to index slightly later stages of processing, perhaps those related

to the integration of the material in the current target region with the preceding text (Rayner,

Warren, Juhasz & Liversedge et al. 2004, Sturt 2003). The fourth measure calculated was

rereading time, the sum of all fixations within the target region after it has been exited to

either the left or the right for the first time. As such, this measure reflects later, second-pass

stages of processing which are somewhat delayed in relation to the measures defined above

that relate to a region’s first encounter. Finally, total viewing time is the summed duration of

all fixations within a target region, irrespective of whether they occurred during its first or

any subsequent passes. As such, this measure is a general index of processing load,

encapsulating both early and later stages of processing.

All measures were calculated contingent on a region gaining a progressive first fixation

during its first-pass. Trials in which a region was initially skipped were treated as missing

data. For rereading time, trials in which a region did not receive any subsequent refixations

following the first-pass were treated as contributing a rereading time of zero to the

calculation of averages. Fixations shorter than 50ms that were within one degree of another

fixation were merged together. All other fixations shorter than 50ms were discarded before

any further analysis, as little information is extracted during such short fixations (Rayner &

Pollatsek 1989). Before any further analysis was undertaken, trials in which track loss

occurred, or in which the target derived word region was directly fixated before reading of

the second sentence had begun, were removed. This amounted to the removal of 1.93% of

the critical trials, equally distributed across conditions.

26

Results

Analysis of the pre-target region is reported first. This region received a progressive first

fixation in 87.41%, 88.36% and 86.84% of trials in the irregular plural, regular plural and

singular conditions respectively. This amounted to an overall first-pass fixation probability

of 87.54%, with no significant differences between conditions (F1(2, 46) < 1, p = .826; F2(2,

51) < 1, p = .857). Mean durations of the five reading time measures calculated for the pre-

target region are shown in Table 2.

//INSERT TABLE 2 ABOUT HERE//

A series of one-way ANOVAs revealed no reliable differences between any of the

conditions, for any of the five reading time measures (all Fs < 1, all ps < .45). This indicates

that there were no reliable effects of the experimental manipulation on the reading times of

the pre-target region, either as a result of parafoveal processing of the derived word before its

initial inspection, or as a result of any subsequent additional processing incurred following

inspection of the derived word.

Consider next the results for the critical word. The derived word region received a

progressive first fixation in 96.95% of trials in the irregular plural condition, 95.06% of trials

in the regular plural condition, and 93.57% of trials in the singular condition. This amounted

to an overall first-pass fixation probability of 95.20%, and a one-way ANOVA revealed

reliable differences in skipping rates between conditions, at least in the subjects analysis

(F1(2, 26) = 3.23, p = .049; F2(2, 51) = 2.10, p = .134). Subsequent pairwise comparisons

indicated no significant differences in first-pass skipping rates between the regular plural and

irregular plural conditions (t1(23) = 1.64, p = .115; t2(34) = 1.38, p = .177), nor between the

regular plural and singular conditions (t1(23) = 1.03, p = .312; t2(34) < 1, p = .439).

27

However, first-pass skipping rates of the derived word were significantly higher in the

singular condition than in the irregular plural condition (t1(23) = 2.36, p = .027; t2(34) = 2.03,

p = .050). Whilst significant, this difference in skipping rates was numerically small (only

3.38%), and most likely not large enough to have any reliable effects on the reading time

measures reported below. Table 3 presents the main eye-movement data.

//INSERT TABLE 3 ABOUT HERE//

A one-way ANOVA revealed a significant main effect of base noun type on the first fixation

duration of the derived word region (F1(2, 46) = 7.79, p = .001; F2(2, 51) = 4.97, p = .011).

Subsequent pairwise comparisons revealed that derived words containing regular plurals had

a significantly longer first fixation duration than derived words containing singular base

nouns (t1(23) = 3.48, p = .002; t2(34) = 2.86, p = .007). Derived words containing irregular

plurals also had a significantly longer first fixation duration than derived words containing

singular bases (t1(23) = 4.23, p < .001; t2(34) = 2.84, p = .008). There were no significant

differences however between the first fixation duration of derived words containing regular

plural base nouns in comparison to those containing irregular plurals (t1(23) < 1, p = .695;

t2(34) < 1, p = .672). The same pattern of results was obtained for the gaze duration of the

derived words. A one-way ANOVA revealed a significant main effect of base noun type

(F1(2, 46) = 7.79, p < .001; F2(2, 51) = 6.09, p = .004), and subsequent pairwise comparisons

revealed significantly longer gaze durations for derived words containing regular plurals in

comparison to those containing singulars (t1(23) = 4.90, p < .001; t2(34) = 3.32, p = .002),

and also significantly longer gaze durations for derived words containing irregular plurals in

comparison to those containing singulars (t1(23) = 5.41, p < .001; t2(34) = 3.13, p = .004).

The gaze durations of derived words containing regular plural bases did not differ

28

significantly from those containing irregular plurals (t1(23) < 1, p = .837; t2(34) < 1, p =

.872).

A one-way ANOVA for the regression path duration revealed a significant main effect of

base noun type in the subjects analysis, that approached significance by items (F1(2, 46) =

5.32, p = .008; F2(2, 51) = 2.57, p = .087). Pairwise comparisons indicated that derived

words containing regular plural base nouns had significantly longer regression path durations

than derived words containing singular bases (t1(23) = 2.68, p = .013; t2(34) = 2.39, p =

.022). Derived words containing irregular plural base nouns also had longer regression path

durations than derived words containing singular bases, though this difference approached

significance in the subjects analysis only (t1(23) = 2.04, p = .053; t2(34) < 1, p = .368).

Finally, there was also a trend in the subjects analysis towards longer regression path

durations for derived words containing regular plural base nouns in comparison to those

containing irregular plurals (t1(23) = 1.79, p = .087; t2(34) = 1.32, p = .197).

For the rereading time data, a one-way ANOVA again revealed a significant main effect of

base noun type (F1(2, 46) = 9.37, p < .001; F2(2, 51) = 10.37, p < .001). Pairwise

comparisons revealed that derived words containing regular plurals had significantly longer

rereading times than both derived words containing singular bases (t1(23) = 3.81, p = .001;

t2(34) = 3.77, p = .001) and derived words containing irregular plurals (t1(23) = 2.96, p =

.007; t2(34) = 4.15, p < .001). There were, however, no significant differences between the

rereading times of derived words containing irregular plurals in comparison to those

containing singular base nouns (t1(23) < 1, p = .760; t2(34) < 1, p = .810).

Finally, a one-way ANOVA of the total viewing time data revealed a significant main effect

of base noun type (F1(2, 46) = 21.36, p < .001; F2(2, 51) = 10.31, p < .001), and subsequent

pairwise comparisons revealed significant differences between each of the three conditions.

Derived words containing irregular plurals had longer total viewing times than derived words

29

containing singular bases (t1(23) = 4.40, p < .001; t2(34) = 1.97, p = .057). Derived words

containing regular plurals also had longer total viewing times than both derived words

containing singular base nouns (t1(23) = 6.42, p < .001; t2(34) = 4.77, p < .001), and derived

words containing irregular plurals (t1(23) = 2.93, p = .008; t2(34) = 2.46, p = .019).

Summarizing the main finding of the present experiment, the experimental manipulation

gave rise to different reading time patterns for the critical derived word region. For the two

reading time measures which reflect processing time incurred during the initial inspection of

the target region (first fixation duration and gaze duration), longer reading times were

obtained for derived words containing plural base nouns, irrespective of regularity, in

comparison to derived words containing singular bases. In these two early first-pass

measures, no reliable differences were obtained between derived words containing regular

plurals in comparison to those containing irregular plurals. By contrast, the one processing

measure which specifically excludes the initial time spent in the derived word region

(rereading time) yielded longer reading times for derived words containing regular plurals in

comparison to those containing either irregular plurals or singulars, with no significant

differences between the latter two. Finally, the two reading time measures which index both

the initial time spent in the target region and any additional subsequent processing incurred

thereafter (regression path duration and total viewing time) yielded a gradual pattern of

responses, with short reading times for derived words containing singular base nouns,

followed by longer reading times for derived words containing irregular plurals, with derived

words containing regular plurals producing the longest reading times.

Discussion

These results indicate that the two constraints we examined affect different stages of

processing. The Category Constraint which permits uninflected and disprefers inflected

30

forms (irrespective of regularity) inside derived forms has an immediate effect on

processing, from the earliest stages of first-pass processing. The Structure Constraint against

regular plurals inside derived words, however, affects reading time measures indexing

subsequent later stages of processing. These differences correspond to different levels of

morphological analysis needed for detecting violations of these two constraints. Category

violations inside derived words, i.e., cases in which a derivational affix is combined with an

inflected word form rather than with a stem, can be immediately recognized once the

derivational suffix is stripped off. By contrast, detecting violations of the constraint against

regular inflections inside derived forms (and compounds) requires a more detailed analysis

of the internal structure of the base noun within a derived word.

Parallel-interactive models of language processing (see e.g. Thornton et al. 1998, Spivey et

al. 2002) claim that the language processor makes use of all relevant constraints in parallel

from the earliest stages of processing. From this perspective, one would have expected that

in all measures of processing, derived words containing singular nouns should yield the

shortest reading times (because these forms do not violate any constraint), followed by

derived words containing irregular plurals (due to the violation of one constraint), with

derived words containing regular plurals yielding the longest reading times (because these

forms violate both constraints). This pattern was indeed found for the offline acceptability

judgments, the total viewing times, and the regression path durations, but importantly, not

for those eye-movement measures that are sensitive to early stages of processing. Thus, the

prediction that all relevant constraints affect processing equally early was clearly

disconfirmed.

Our results appear to be more consistent with stage-based of models of language processing.

For concreteness, consider Friederici’s (2002) three-stage model of sentence processing

which distinguishes between an early stage (phase I), when the processing system identifies

31

word-category information (e.g. determiner [DET], noun [N], verb[V] etc.), phase II at

which point syntactic dependencies and other aspects of morpho-syntactic structure are

computed, and a late phase III that comprises of sentence-level integration and reanalysis

processes. According to Friederici (2002), results from event-related brain potentials (ERPs)

indicate that each stage is associated with a distinct ERP component, phase I with the so-

called Early Anterior Negativity (ELAN) between 150 and 200 ms after stimulus onset for

rapidly detectable word-category violations, phase II with a Left Anterior Negativity (LAN)

between 300 and 500 ms for other kinds of morphosyntactic error, and phase III with a late

positivity (P600) around 600 ms after stimulus onset for syntactic anomalies such as garden-

path sentences that require syntactic reanalysis. In addition, Grodzinsky & Friederici (2006)

identified selective activations of different cortical areas for these stages of sentence

processing by using data from neuroimaging studies. Extending this model to the domain of

morphology, one would expect that violations of the Category Constraint (against inflected

words) are rapidly detected at an early stage of processing and that violations of the Structure

Constraint (against regular inflections) are recognized later. Our results confirm this

prediction and suggest that the processing of a derived form involves two distinct stages, an

initial stage when the derivational affix is stripped off, yielding a simple [[base]+affix]

structure, and a later stage in which the internal structure of the base is analyzed. The

Category Constraint appears to affect the initial stage of processing derived words, whereas

violations of the Structure Constraint are detected later, once the internal structure of the

base has been analyzed.

The proposed account of the eye-movement data hinges on the claim that the dispreference

of regular and irregular plural forms inside derivations is due to the morphological category

of these forms and that regular and irregular plural forms are inflected words rather

uninflected bases. Alternatively, however, we may say that the contrast between singular and

32

plural forms inside derivations is due to the same semantic constraint (against plural

meaning) that has been proposed for compounds (Haskell et al. 2003). Clearly, the effect of

this semantic constraint would be identical to that of our Category Constraint, in that both

constraints prevent regular and irregular plural forms from occurring inside compounds or

derived forms. Whilst it is indeed difficult to distinguish between effects of these two

constraints in English (due to the fact that a singular noun is identical in form to an

uninflected base), we argued that, due to linguistic differences between compounding and

derivation, the semantic constraint does not apply to derivational forms (see section 4

above). For example, the kinds of derivational affixes we tested do not have referents, and

the base of a derived word does not constitute a ‘modifier’. Hence, a constraint that specifies

the semantic relationship between the modifier and the head of a compound does not apply

to derived words. Instead, derivational affixes are subject to selectional restrictions, and the

Category Constraint captures one such restriction, namely that inflected words should not

feed derivational processes.

Further experimental research is needed to determine whether the findings from the eye-

movement experiment reported above can be replicated and the suggested interpretation be

maintained. It would also be useful to examine the processing of derivational forms and their

interaction with inflectional processes in languages in which it is possible to distinguish

more clearly between inflected singular forms and uninflected bases than in English. Another

promising avenue for further research would be to employ the ERP methodology for

investigating these kinds of phenomena. Like eye-movement monitoring, ERPs provide

excellent measures of the time-course of language processing. In addition, distinct ERP

effects have been found for lexical-semantic violations (N400) and morpho-syntactic

violations (ELAN, LAN P600). Thus, data from ERP experiments may provide crucial

33

evidence on how different types of information affect the on-line processing of derivational

forms and other morphologically complex words.

6. Conclusion

The two experiments reported in the present paper each revealed a preference for derived

words containing singular base nouns. In Experiment 1 (as well as in the pre-test of the

materials for experiment 2), derived words containing plural bases received lower

acceptability ratings than derived words containing singular bases. Moreover, in Experiment

2, derived words containing plurals received, for example, longer total viewing times than

those containing singular bases. These findings can be interpreted as resulting from a

constraint of derived word formation (labeled the Category Constraint) according to which

inflected words are dispreferred inside derivational forms. Additionally, the two experiments

revealed that regular plurals are less acceptable than irregular plurals inside derived words.

Derived words containing regular plural bases received lower acceptability ratings than those

containing irregular plurals in Experiment 1, and derived words containing regular plurals

received, for example, longer total viewing times than those containing irregular plurals in

Experiment 2. The results of Experiment 1 (see ‘item set 2’) confirmed that the dissociation

between regular and irregular plurals inside derived words cannot be explained in

phonological terms. Instead, these findings were interpreted as resulting from a

morphological constraint (labeled the Structure Constraint) that prevents regularly inflected

forms (which are morphologically structured and decomposable) from feeding any further

word-formation processes.

We found that these two constraints have distinct time-courses during morphological

processing. The two first-pass measures examined in Experiment 2 (First fixation durations,

gaze duration) showed that derived words containing either regular or irregular plural base

34

nouns had equally longer reading times than derived words containing singular nouns,

indicating that violations of the Category Constraint have an immediate effect on processing.

The later measures, on the other hand, revealed a contrast between regular and irregular

plurals. Derived words containing regular plurals had longer rereading times and longer total

viewing times than derived words containing irregular plurals, indicating that violations of

the Structure Constraint are recognized later. Taken together, these findings are compatible

with stage-based models of language processing and specifically with the hypothesis that

category information is processed earlier than other aspects of morpho-syntactic structure

(Friederici 2002, Grodzinsky & Friederici 2006).

Acknowledgements

Supported by an ESRC postgraduate training scholarship. We are grateful to Iris Berent,

Keith Rayner, the editors of The Mental Lexicon, and the members of the Psycholinguistics

Research Group at our department for comments and helpful discussion.

35

Footnotes

1 Concreteness and familiarity ratings were collected for the singular forms of each noun,

as the MRC database did not provide this information for the plural forms. Concreteness

and familiarity ratings were not available for the singular forms of two items in the

irregular plural condition (louse and goose), nor was a concreteness rating available for

the singular form of one item in the regular plural condition (son).

2 Lexical decision and naming latencies were available for all items except the regular

plural noun mites.

Table 1: Mean acceptability ratings (and standard deviations) for derived forms

Item Set 1 Item Set 2

IRREGULAR REGULAR phon/SIM phon./DIS

singular 5.15 (1.22) 5.07 (1.12 5.08 (1.02) 5.07 (1.09)

plural 4.12 (1.35) 2.87 (1.23)

37

Table 2: Mean durations (and standard deviations) of five reading time measures for the pre-

target region

Irregular Plural Regular Plural Singular

First Fixation Duration 209 (23) 209 (28) 209 (25)

Gaze Duration 233 (39) 238 (47) 229 (36)

Regression Path Duration 288 (73) 308 (114) 308 (79)

Rereading Time 79 (54) 94 (54) 90 (64)

Total Viewing Time 312 (71) 331 (88) 319 (82)

38

Table 3: Mean durations (and standard deviations) of five reading time measures for the

derived word region

Irregular Plural Regular Plural Singular

First Fixation Duration 258 (35) 255 (52) 237 (42)

Gaze Duration 366 (120) 363 (137) 297 (90)

Regression Path Duration 423 (124) 467 (186) 389 (114)

Rereading Time 126 (81) 224 (168) 120 (81)

Total Viewing Time 492 (131) 587 (210) 417 (143)

Appendix A: Frequencies and length of base nouns used in Item Set 1 of Experiment 1

Irregular Regular

Frequency Letters Syllables

Frequency Letters Syllables

Louse 1 5 1 Flea 4 4 1

Ox 4 2 1 Pig 43 3 1

Goose 6 5 1 Duck 14 4 1

Mouse 8 5 1 Rat 24 3 1

Tooth 13 5 1 Eye 523 3 1

Foot 98 4 1 Hand 724 4 1

Woman 338 5 2 Sister 114 6 2

Child 426 5 1 Baby 258 4 2

Man 975 3 1 Brother 138 7 2

Average 207.67 4.33 1.11 204.67 4.22 1.33

40

Appendix B: Frequencies and length of base nouns used in Item Set 2 of Experiment 1

Phonologically Similar Phonologically Dissimilar

Frequency Letters Syllables

Frequency Letters Syllables

Box 102 3 1 Crate 6 4 1

Axe 9 3 1 Spear 12 3 1

Fox 15 3 1 Wolf 10 4 1

Prize 26 5 1 Award 17 3 2

Rose 21 4 1 Daisy 32 3 2

Blouse 11 6 1 Hat 68 4 1

Maize 6 5 1 Corn 24 6 1

Nose 81 4 1 Ear 88 4 1

Average 33.88 4.13 1.00 32.13 4.25 1.25

Appendix C: Frequencies and length of base nouns used in Experiment 2

Irregular Plural Regular Plural Singular

Frequency Letters Syllables

Frequency Letters Syllables

Frequency Letters Syllables

Lice 2 4 1 Mites 2 5 1 Flea 4 4 1

Oxen 2 4 2 Animals 260 7 3 Horse 132 5 1

Geese 5 5 1 Birds 103 5 1 Fish 163 4 1

Mice 10 4 1 Cats 67 4 1 Dog 115 3 1

Teeth 75 5 1 Eyes 523 4 1 Face 460 4 1

Feet 229 4 1 Arms 210 4 1 Hand 724 4 1

Women 511 5 2 Girls 438 5 1 Lady 112 4 2

Children 655 8 2 Boys 349 4 1 Baby 258 4 2

Men 655 3 1 Sons 206 4 1 Husband 154 7 2

Average 238.22 4.67 1.33 239.78 4.67 1.22 235.78 4.33 1.33

References

Alegre, M. and Gordon, P. (1996). Red rats eater exposes recursion in children’s word

formation. Cognition 60: 65-82.

Alegre, M. and Gordon, P. (1999). Why parks commissioners aren’t rats eaters: Semantic

constraints on regular plurals inside compounds. Unpublished manuscript.

Altmann, G., Garnham, A., & Dennis, Y. (1992). Avoiding the garden path: Eye movements

in context. Journal of Memory and Language 31: 685-712.

Anderson, S. (1992). A-morphous morphology. Cambridge: Cambridge University Press.

Aronoff, M. (1976). Word formation in generative grammar. Cambridge, MA: MIT Press.

Baayen, H., Piepenbrock, R. & van Rijn, H. (1993). The CELEX lexical database (CD

ROM). Philadelphia, PE: Linguistic Data Consortium, University of Pennsylvania.

Balota, D., Cortese, M., Hutchison, K., Neely, J., Nelson, D., Simpson, G. and Treiman, R.

(2002). The English lexicon project: A web-based repository of descriptive and

behavioral measures for 40,481 English words and nonwords. http://elexicon.wustl.edu/,

Washington University.

Berent, I. & Pinker, S. (2007). The dislike of regular plurals in compounds: Phonological

familiarity or morphological constraint? The Mental Lexicon 2: 129-181.

Binder, K., Duffy S. & Rayner, K. (2001). The effects of thematic fit and discourse context

on syntactic ambiguity resolution. Journal of Memory and Language 44: 297-324.

Borer, H. (1988). On the parallelism between compounds and constructs. Yearbook of

Morphology 1: 45-66.

Cunnings, I. & Clahsen, H. (2007). The time-course of morphological constraints: Evidence

from eye-movements during reading. Cognition 104: 476-494.

43

Di Sciullo, A., and Williams, E. (1987). On the definition of word. Cambridge, MA: MIT

Press.

Fabb, N (1988). English suffixation is constrained only by selectional restrictions. Natural

Language and Linguistic Theory 6: 527-539.

Frazier, L. & Fodor, J. D. (1978). The sausage machine: A two-stage parsing model.

Cognition 6: 291-325.

Frazier, L. & Clifton, C. (1996). Construal. Cambridge, MA: MIT Press.

Friederici, A. (2002). Towards a neural basis of auditory sentence processing. Trends in

Cognitive Sciences 6: 78-84.

Gordon, P. (1985). Level-ordering in lexical development. Cognition 21: 73-93.

Greenberg, J. (1966), Some universals of grammar with particular reference to the order of

meaningful elements. In Greenberg, J. (ed.), Universals of language. 2nd edition.

Cambridge, MA: MIT Press, 73-113.

Grodzinsky, Y. & Friederici, A. (2006). Neuroimaging of syntax and syntactic processing.

Current Opinion in Neurobiology 16: 240-246.

Haskell, T., MacDonald, M. & Seidenberg, M. (2003). Language learning and innateness:

Some implications of compounds research. Cognitive Psychology 47: 119-163.

Hay, J. (2002). From speech perception to morphology: Affix ordering revisited. Language

78: 527-555.

Hay, J. (2003). Causes and consequences of word structure. New York: Routledge.

Juhasz, B. and Rayner, K. (2003). Investigating the effects of a set of intercorrelated

variables on eye fixation durations in reading. Journal of Experimental Psychology:

Learning, Memory and Cognition 29: 1312-1318.

44

Kiparsky, P (1982). From cyclic phonology to lexical phonology. In van der Hulst, H. &

Smith, N. (eds.), The structure of phonological representations (Part 1). Dordrecht:

Foris, 131-175.

Marshall, C. & van der Lely, H. (2007). Derivational morphology in children with

grammatical-specific language impairment. Clinical Linguistics & Phonetics 21: 71-91.

Matthews, P. (1991). Morphology. 2nd edition. Cambridge, Cambridge University Press.

McElree, B. & Griffith, T. (1998). Structural and lexical constraints on filling gaps during

sentence comprehension: A time-course analysis. Journal of Experimental Psychology:

Learning, Memory, and Cognition 24: 432-460.

Pinker, S. (1999). Words and rules. New York: Basic Books.

Plag, I. (1996). Selectional restrictions in English suffixation revisited. A reply to Fabb

(1988). Linguistics 34: 769-798.

Plag, I. (1999). Morphological productivity. Structural constraints in English derivation.

Berlin: Mouton de Gruyter.

Plag, I. (2002). The role of selectional restrictions, phonotactics and parsing in constraining

suffix ordering in English. Yearbook of Morphology 2001: 285- 314.

Rayner, K., (1998). Eye movements in reading and information processing: 20 years of

research. Psychological Bulletin 124: 372-422.

Rayner, K. Carlson, M. & Frazier, L. (1983). The interaction of syntax and semantics during

sentence processing: Eye movements in the analysis of semantically biased sentences.

Journal of Verbal Learning and Verbal Behavior 22: 358-374.

Rayner, K. & Pollatsek, A. (1989). The psychology of reading. New Jersey: Prentice Hall.

45

Rayner, K., Warren, T., Juhasz, B. & Liversedge, S. (2004). The effect of plausibility on eye

movements in reading. Journal of Experimental Psychology: Learning, Memory and

Cognition 30: 1290-1301.

Senghas, A., Kim, J. & Pinker, S. (2005). The plurals-in-compounds effect. Unpublished

Manuscript.

Sonnenstuhl, I. & Huth, A. (2002). Processing and representation of German -n plurals: A

dual mechanism approach. Brain & Language 81: 276-290.

Spivey, M. J, Fitneva, S. A., Tabor, W. & Ajmani, S. (2002). The time course of information

integration in sentence processing. In Merlo, P. & Stevenson, S. (eds.), The lexical basis

of sentence processing. Amsterdam: John Benjamins, 207-232.

Stump, G. (2001). Inflectional morphology: A theory of paradigm structure. Cambridge:

Cambridge University Press.

Sturt, P. (2003). The time-course of the application of binding constraints in reference

resolution. Journal of Memory and Language 48: 542-562.

Thornton, R., Gil, M., & MacDonald, M.C. (1998). Accounting for cross-linguistic variation:

A constraint-based perspective. In Hillert, D (ed.), Syntax and semantics vol. 31: A

crosslinguistic perspective. San Diego: Academic Press.

Wiese, R. (1996). The phonology of German. Oxford: Oxford University Press.

Wilson, M. (1988). The MRC Psycholinguistic Database: Machine Readable Dictionary,

Version 2. Behavioral Research Methods 20: 6-11.